Detecção automática da superfície sagital mediana em imagens de ressonância magnética do cérebro humano usando uma abordagem de filtragem multiescala

• Ano de defesa: 2020
• Autor(a) principal: Frascá, Fernando de Natali
• Orientador(a): Ferrari, Ricardo José
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de São Carlos; Câmpus São Carlos
• Programa de Pós-Graduação: Programa de Pós-Graduação em Ciência da Computação - PPGCC
• Palavras-chave em Português: Imagens de ressonância magnética; Segmentação de imagens médicas; Superfície sagital mediana; Plano sagital mediano; Segmentação automática
• Palavras-chave em Inglês: Magnetic resonance images; Medical images segmentation; Mid-sagittal surface; Mid-sagittal plane; Automated segmentation
• Área do conhecimento CNPq: CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO
• Link de acesso: https://repositorio.ufscar.br/handle/ufscar/13086

Both hemispheres of the human brain, left and right, are separated by a fissure called interhemispheric (IHF), which is commonly shaped by a geometric plane, known as midsagittal plane (MSP). However, despite the name, the MSP is not always plane, once the brain is not perfectly symmetric. The detection of the MSP in human brain images is an essential task to segment the brain hemispheres. Studies suggest that abnormal values of asymmetrical brain may be related to traumas and neurological diseases. Through the years, various computer methods were proposed to detect MSP in an automated way, nonetheless the main methods that were found in literature focus on the detection of a plan, imposing such a geometric restriction on IHF. This work proposes a new automatic computational technique for the detection of a median sagittal surface (SSM) in magnetic resonance (MR) images for application in the analysis of asymmetries between cerebral hemispheres. The proposed method uses a MSP as guidance to the definition of an region of interest (ROI) that potentially contains an IHF. The RM 3D image is decomposed in 2D slices, which are processed, filtered and piled to form an MSS. Based on the results obtained, it appears that the proposed method enables a higher precision when getting the volumetric measures and therefore in the calculation of brain asymmetries. The MSS precision is evaluated comparing the left and right hemispheres segmented from the detected MSS, with the sketches handmade by two specialists.